Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti, Robert West
{"title":"On the conversational persuasiveness of GPT-4","authors":"Francesco Salvi, Manoel Horta Ribeiro, Riccardo Gallotti, Robert West","doi":"10.1038/s41562-025-02194-6","DOIUrl":null,"url":null,"abstract":"<p>Early work has found that large language models (LLMs) can generate persuasive content. However, evidence on whether they can also personalize arguments to individual attributes remains limited, despite being crucial for assessing misuse. This preregistered study examines AI-driven persuasion in a controlled setting, where participants engaged in short multiround debates. Participants were randomly assigned to 1 of 12 conditions in a 2 × 2 × 3 design: (1) human or GPT-4 debate opponent; (2) opponent with or without access to sociodemographic participant data; (3) debate topic of low, medium or high opinion strength. In debate pairs where AI and humans were not equally persuasive, GPT-4 with personalization was more persuasive 64.4% of the time (81.2% relative increase in odds of higher post-debate agreement; 95% confidence interval [+26.0%, +160.7%], <i>P</i> < 0.01; <i>N</i> = 900). Our findings highlight the power of LLM-based persuasion and have implications for the governance and design of online platforms.</p>","PeriodicalId":19074,"journal":{"name":"Nature Human Behaviour","volume":"40 1","pages":""},"PeriodicalIF":21.4000,"publicationDate":"2025-05-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Nature Human Behaviour","FirstCategoryId":"102","ListUrlMain":"https://doi.org/10.1038/s41562-025-02194-6","RegionNum":1,"RegionCategory":"心理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
引用次数: 0
Abstract
Early work has found that large language models (LLMs) can generate persuasive content. However, evidence on whether they can also personalize arguments to individual attributes remains limited, despite being crucial for assessing misuse. This preregistered study examines AI-driven persuasion in a controlled setting, where participants engaged in short multiround debates. Participants were randomly assigned to 1 of 12 conditions in a 2 × 2 × 3 design: (1) human or GPT-4 debate opponent; (2) opponent with or without access to sociodemographic participant data; (3) debate topic of low, medium or high opinion strength. In debate pairs where AI and humans were not equally persuasive, GPT-4 with personalization was more persuasive 64.4% of the time (81.2% relative increase in odds of higher post-debate agreement; 95% confidence interval [+26.0%, +160.7%], P < 0.01; N = 900). Our findings highlight the power of LLM-based persuasion and have implications for the governance and design of online platforms.
期刊介绍:
Nature Human Behaviour is a journal that focuses on publishing research of outstanding significance into any aspect of human behavior.The research can cover various areas such as psychological, biological, and social bases of human behavior.It also includes the study of origins, development, and disorders related to human behavior.The primary aim of the journal is to increase the visibility of research in the field and enhance its societal reach and impact.